Evidence From The Housing Market in New York City
Old Dominion University
via The Wall Street Journal
The Police Department has failed to bring down crime. Perpetrators of violent crime in New York City are concentrated among relatively small groups in areas with high rates of crime. The Police Department has sophisticated databases to help it understand who they are. It has to make more effective use of “precision policing,” including the narrow and tactical use of “stop and frisk” to focus its resources on those who are thought to be the primary drivers of violence.
NYT (2024)
How do we value the way we are policed?
Homeowner distaste for police hostility, or the “reminder” of crime, may reduce housing prices (especially if the crime benefits are minimal).
Even if Stop & Frisk may do little to prevent crime, it could give off the perception of safety.
This is an important, timely question as cities contemplate different policing strategies in the wake of violence, riots, etc.
MacDonald, Fagan, and Geller (2016)
In January 2003, the NYPD deployed roughly two-thirds of its police academy graduates—about 1,500 new police officers—to Impact Zones.
I want to know how property prices changes after Stop & Frisk was ruled unconstitutional.
Idea 1: Calculate Average Prices Before & After
This won’t work because there might be price dynamics going on in NYC that occur at the same time as the Floyd case. How can I be sure I am not capturing those effects?
I want to know how property prices changes after Stop & Frisk was ruled unconstitutional.
Idea 1: Calculate Average Prices Before & After
This won’t work because there might be price dynamics going on in NYC that occur at the same time as the Floyd case. How can I be sure I am not capturing those effects?
Idea 2: Calculate Difference in Prices between Treatment and Control
This won’t work because there might be pre-existing differences between treatment and control. How can I be sure I am not capturing those effects?
I want to know how property prices changes after Stop & Frisk was ruled unconstitutional.
Idea 1: Calculate Average Prices Before & After
This won’t work because there might be price dynamics going on in NYC that occur at the same time as the Floyd case. How can I be sure I am not capturing those effects?
Idea 2: Calculate Difference in Prices between Treatment and Control
This won’t work because there might be pre-existing differences between treatment and control. How can I be sure I am not capturing those effects?
Idea 3: Combine Ideas 1 and 2
Calculate difference (before and after) for both treatment and control. Whatever happened in the control should have also happened in the treatment. Subtracting the change in the control from the change in the treatment will isolate the treatment effect.
I want to know how property prices changes after Stop & Frisk was ruled unconstitutional.
Idea 1: Calculate Average Prices Before & After
This won’t work because there might be price dynamics going on in NYC that occur at the same time as the Floyd case. How can I be sure I am not capturing those effects?
Idea 2: Calculate Difference in Prices between Treatment and Control
This won’t work because there might be pre-existing differences between treatment and control. How can I be sure I am not capturing those effects?
Idea 3: Combine Ideas 1 and 2
Calculate difference (before and after) for both treatment and control. Whatever happened in the control should have also happened in the treatment. Subtracting the change in the control from the change in the treatment will isolate the treatment effect. This is called difference-in-differences.
\[\begin{align}\text{log}(P_{it}) =& \ X_{it}'\beta + \delta_1 \text{Floyd}_{t} + \delta_2 \text{Impact}_{i} \\ &+ \delta_3 (\text{Floyd}_{t} \times \text{Impact}_{i}) + \epsilon_{it}\end{align}\]
\(\delta_3\) is the “treatment effect”: how much more did properties prices change in Impact Zones, relative to nearby properties, when Stop & Frisk was ruled unconstitutional.
If property prices for both areas changed in similar ways, \(\delta_3 \approx 0\).
We replicate and “extend” the analysis in Friedman (2015):
Virginia Polytechnic Institute and State University